Weibull hazard rate
Hazard rate h(t) is the intensity of failure at age t, given survival to t. Reliability work uses it to distinguish infant mortality, useful life, and wear-out. Iron Lake Reliability evaluates h(t) from the fitted LifeFit; this page states the 2P Weibull formula and a numerical check.
Open the Iron Lake Reliability workbook Open Weibull Calculator Evaluate h(t) from a fit
Formula
With density f(t) = (β/η)(t/η)β−1 exp[−(t/η)β] and reliability R(t) = exp[−(t/η)β],
h(t) = f(t) / R(t) = (β/η) (t/η)β−1
- β < 1 — decreasing hazard (DFR), infant mortality, burn-in candidates.
- β = 1 — constant hazard, exponential, h = 1/η = 1/MTTF.
- β > 1 — increasing hazard (IFR), wear-out, replacement / inspection age matters.
A single two-parameter Weibull is monotone. The decorative bathtub on the Iron Lake Reliability header is three example shapes overlaid, not a fitted mixture.
Worked example (β = 2, η = 1 000 h)
At t = 500 h:
h(500) = (2 / 1 000) × (500 / 1 000)1 = 0.002 × 0.5 = 0.001 / h
That is one expected failure per 1 000 surviving unit-hours at age 500 h. At t = η = 1 000 h the same model gives h = 0.002 / h — twice as steep, because β = 2 is linear wear-out in t. For β = 1, η = 1 000 h, h would stay 0.001 / h at every age.
| t (h) | R(t) | h(t) (1/h) |
|---|---|---|
| 250 | 0.939 | 0.0005 |
| 500 | 0.779 | 0.0010 |
| 1 000 | 0.368 | 0.0020 |
Cumulative hazard, and why h(t) is not a probability
The exponent in R(t) is the cumulative hazard, H(t) = (t/η)β, and R(t) = exp[−H(t)]. For the worked example, H(500) = 0.52 = 0.25 and R(500) = exp(−0.25) = 0.778801, the 0.779 in the table. H is dimensionless. h is a rate per life-unit. A value of 0.001 / h means one expected failure per 1 000 hours among units still alive at that age. It is not a failed fraction. At the same age F(500) = 0.2212, so just over 22% have already failed. Those are different statements.
Over a short interval the conditional failure probability, given survival to t, is about h(t) · Δt, and the approximation holds when that product is much less than 1. At 500 h, h · 1 h = 0.001. The exact probability of failing before 501 h, given survival to 500 h, is 1 − R(501) / R(500) = 0.0010005. Rate and one-hour probability agree because 0.001 is small. They do not agree over a long window: h(500) × 500 h = 0.5, which is not a probability to quote. Use F(t), or 1 − R(t + Δ) / R(t), for a probability.
Same scale, different shape
Hold η = 1 000 h and compare hazards. For β = 2, h(t) = 2t / η2, so h(500) = 1 000 / 1 000 000 = 0.001 / h. For β = 1, h = 1/η = 0.001 / h at every age. The match at 500 h is exact: β (t/η)β−1 = 1 when β = 2 and t = η/2. Before that age the wear-out hazard is lower than the exponential; after it, higher. From 250 h to 1 000 h the β = 2 hazard rises from 0.0005 / h to 0.002 / h, fourfold, because a shape of 2 is linear in t.
| β | h(250 h) | h(500 h) | h(1 000 h) | Direction |
|---|---|---|---|---|
| 0.5 | 0.001000 / h | 0.000707 / h | 0.000500 / h | falling |
| 0.8 | 0.001056 / h | 0.000919 / h | 0.000800 / h | falling |
| 1 | 0.001 / h | 0.001 / h | 0.001 / h | flat |
| 2 | 0.0005 / h | 0.001 / h | 0.002 / h | rising, linear in t |
| 3 | 0.0001875 / h | 0.00075 / h | 0.003 / h | rising, proportional to t2 |
A higher β is not a higher hazard at every age. The flat row is the only one in the table where h(t) equals 1/MTTF at every age. For β = 2, 1 / 886.227 h = 0.001128 / h, near h(500) and far from h(1 000).
Wear-out in one comparison
Conditional reliability R(T + Δ) / R(T) is the chance of surviving an extra Δ given survival to T. For β = 2 and η = 1 000 h, a new unit has R(200) = exp(−0.04) = 0.9608. A unit already 500 h old has R(700) / R(500) = 0.612626 / 0.778801 = 0.7866 over the next 200 h. The same ratio for β = 1 is exp(−0.2) = 0.8187, and R(200) equals it: age does not matter. A single Weibull cannot draw a bathtub, because its hazard never turns around.
When to use h(t), and common mistakes
Use h(t) when the question is about age, and F(t) or a B-life when it is a failed fraction. Fit on the Weibull analysis page before quoting a rate.
- Calling 0.001 / h a 0.1% failure probability. Convert with h(t) · Δt only for a short Δt, and say so.
- Replacing a rising hazard with 1/MTTF for the whole life. For this β = 2 model that single number cannot describe both 250 h and 1 000 h.
- Comparing a per-hour hazard with a per-cycle hazard. Convert the time base with a duty cycle you measured.
- Describing a bathtub with one β. If early failures and later wear-out are both in the data, one monotone hazard is the wrong summary. The tutorial is where mixed life histories are handled as a modeling choice.
Fit data on the workbook, then Calculator → Failure rate. Closed-form B10 and MTTF for this same (β, η) are on the B10 and MTTF pages. A full life-data walk-through is in the tutorial.
FAQ
What is the hazard rate h(t)?
h(t) = f(t) / R(t) is the instantaneous failure rate at age t among units that have survived so far. It is not the fraction failed F(t), and it is not 1/MTTF except for the exponential.
What is the Weibull hazard formula?
h(t) = (β/η) (t/η)^{β−1} for t > 0. If β = 1 the power vanishes and h(t) = 1/η (constant). If β > 1 the hazard rises (wear-out). If β < 1 it falls (infant mortality).
How does this relate to the bathtub curve?
A bathtub is often three Weibull (or mixed) regimes in series: β < 1 early, β ≈ 1 in useful life, β > 1 in wear-out. A single 2P Weibull has a monotone hazard — it cannot draw a full bathtub by itself.
Where is h(t) in Iron Lake Reliability?
Weibull Calculator → Rate → Failure rate h(t). Enter the evaluation time in the same life unit as the workbook. The header bathtub artwork is decorative, not a fitted hazard plot.
Open the Iron Lake Reliability workbook Open Weibull Calculator